> So although `read_minibuf' does call `add-to-history' before returning
> (and after setting the current buffer to be the minibuffer's), my input
> wasn't added to any list (that I know of). This is strange, since
> `add-to-history' seemed to work correctly with buffer-local variables
> when I tested it separately. I would've expected the new history item to
> be added to one of the minibuffer's `test-history' local variable.
>
> So I think the first step would be to make `read_minibuf' add the new
> history item into the right place when HIST is buffer-local.

Turns out `get_minibuffer' sets the minibuffer's buffer mode to
`minibuffer-inactive-mode', which resets the minibuffer's local
variables (though not 100% sure about this). I think the history items
were being added to the minibuffer's local version of the variable, and
then being deleted by that.

I created a small patch that sets back the original buffer at the end of
`read_minibuf', so that the history item is added in the right
place. Then, I modified `goto-history-element' as Michael suggested, so
that the function reads the buffer-local version of the variable, but
not the minibuffer's (and updated `minibuffer-history-isearch-wrap' just
in case as well). With this, M-p and M-n works again.

I'm attaching a draft patch in case anyone wants to provide some
feedback. Thanks!

- Fede